Feasibility of Atrial Activation Time ImagingF Hanser1,2, B Tilg1, R Modre1, G Fischer1, B Messnarz1,2, F Hintringer3, FX Roithinger31Institute for Medical Signal Processing and Imaging,University for Health Informatics and Technology Tyrol, Innsbruck, Austria2Institute of Biomedical Engineering, Graz University of Technology, Austria3Clinical Department of Cardiology, University Hospital Innsbruck, AustriaAbstractThe feasibility of atrial activation time imaging isinvestigated based on data sets of four patients whounderwent an electrophysiologic study. Several pacingprotocols with pacing sites at the right atrial appendage,coronary sinus, and high right atrium were part of thestudy and were employed to reconstruct the associatedatrial activation time patterns. The localization error wasestimated to be between 8 and 14 mm.1. IntroductionAtrial and ventricular surface activation time imagingfrom body-surface ECG mapping data is supposedto become a diagnostically powerful clinical tool forassessing cardiac arrhythmias [1]. This cardiac sourceimaging technique aims at providing noninvasivelyinformation about electrical excitation in order to assist thecardiologist in developing strategies for the treatment ofcardiac arrhythmias. Common cardiac arrhythmias, suchas the Wolf-Parkinson-White syndrome, atrioventricularnodal reentrant tachycardia, or atrial fibrillation, can, atleast in many cases, be traced back to accessory pathways,atrial or ventricular foci, e.g. from the pulmonary veins[2, 3], and reentrant circuits. Identifying the site of originof the ectopic focus or the location of an accessory pathwayprovides the essential information for treatment strategies,such as catheter ablation [4].Activation time imaging from 3D+time anatomical andbody-surface ECG mapping data enables noninvasivelythe imaging of the electrical function in the heart [5].The method yields solutions to the electrocardiographicinverse problem and is based on an electrodynamic modelof the patient’s volume conductor and heart. The modelof the heart includes a model of both the atrium andventricle. A separate model for the atrium and ventricle hasbeen inevitable because whole heart models still resist acomputational and technical implementation for providingsolutions to the electrocardiographic inverse problem.Atrial activation time imaging constitutes a moresophisticated problem with respect to ventricular activationtime imaging. The complex geometry of the atria (orificesof the pulmonary veins, orifices of superior and inferiorvena cava, tricuspid and mitral annuli, and right andleft appendages) makes it more difficult to generate ageometrical model that qualifies for a boundary elementformulation. The poor contrast in the MR images makesthe segmentation process quite delicate. In addition, onlythe endocardial boundaries can be seen in the atrial MRimages. Epicardial boundaries have, therefore, to beconstructed artificially. A signal-to-noise ratio of about20dB (more than 40dB in ventricular activation timeimaging) and the significantly smaller effective rank ofthe ECG data matrix impose additional challenges on thestability of the inverse algorithm.In this work we investigate the feasibility of atrialactivation time imaging based on data sets of four patientswho underwent an electrophysiologic study.2. Data acquisition and modelingBody-surface ECG data were acquired under clinicalconditions with a 62-lead ECG mapping system. TheMark-8 body-surface potential mapping system (BiosemiV.O.F., The Netherlands) is an on-line portable computeracquisition system with data transmission via optical fiber.A radiotransparent carbon electrode array was utilized torecord unipolar ECG signals from 62 torso sites (anterior41, posterior 21) while simultaneously enabling X-rayexamination. The Wilson Central Terminal defined, asusual, the reference potential [6]. Electrode signals wereamplified and high-pass and low-pass filtered at edgefrequencies of 0.3Hz and 400Hz with a first and fourthorder analog Bessel filter, respectively. Analog-to-digitalconversion was realized by a 16 bit AD converter at asampling rate of 2048Hz per channel and a quantizationresolution of 500nV /bit. No additional digital filteringwas applied.The torso was imaged with a 1.5Tesla MR scanner0276−6547/02 $17.00 © 2002 IEEE 601 Computers in Cardiology 2002;29:601−604.(Siemens Vision Plus). The myocardium was additionallyimaged in an ECG-gated breath-hold oblique imagingmode in order to model the heart’s surface. Vitamin Emarkers were utilized to determine 7 reference positionsfrom the axial MR scans to be able to couple MRIwith ECG data. The 62 electrode and 7 referencepositions were acquired with the FASTRAK system(Polhemus Inc., USA). The entire volume conductorsincluding the blood masses were modeled with about 4100triangles [7]. The atrial surfaces were represented byabout 1050 triangular elements. The different individualcompartments comprised the torso, the lungs, and theblood mass. The associated conductivities were assumedto be 0.2, 0.08, and 0.6Sm−1, respectively. Figure 1shows a geometrical representation of a patient’s volumeconductor together with electrode positions of the 62-leadECG mapping system.Figure 1. Geometrical model of an individual patient’storso. The model depicts all electrodynamically relevantcompartments: torso, lungs, and blood masses. Spheresindicate electrode positions of the 62-lead ECG mappingsystem.3. Inverse methodEndocardial and epicardial activation time patterns werereconstructed by employing a method capable of providingsolutions to the electrocardiographic inverse problem.Mathematically speaking, the relation between activationtime on the atrial surface and the body surface ECG isrepresented by a nonlinear inverse ill-posed problem. Itcan be formulated in the following form:Fτ = D, (1)where F is a nonlinear operator which maps the activationtime τ onto the body surface ECG data D. Assuming{τk}∞k=0to be a series of approximations of the truesolution τ and linearizing Eq. (1) in the point τkyieldsFτk+ Fk(τ − τk) = D, (2)where Fkis an abbreviation for F (τk) and F representsthe Frechet derivative of the operator F. Equation (1) canbe written in a technically more useful formFkτ = Dk, (3)with Dk= D + Fkτk− Fτk. Equation (3) is, ingeneral, again ill-posed. In order to find a regularizedapproximation for τ a regularization method for linear ill-posed problems can be employed. Applying
View Full Document
Unlocking...